Total factor productivity, health and spatial dependence in some European regions.
Alexa, Diana ; Pop-Silaghi, Monica ; Cismas, Laura Mariana 等
INTRODUCTION
The acceleration of the European Union (EU) integration process not
only brought along the benefits that come with an enlarged common
economic market, but has also enhanced the existing disparities between
the European regions. These disparities, although more pronounced in the
new member states, are an ongoing concern for all EU members. This is
why the main objective of the regional policy of EU is to reduce
disparities within European Community and encourage the development and
convergence at the regional level. Also, the policy mix at the Community
level aims to stimulate competitiveness and boost productivity growth
especially through innovation. The Lisbon Strategy, followed by the 2020
Agenda, promotes a sustainable growth model that is based on
productivity growth and innovation, while also advocating for social
inclusion, increases in human capital and greater concern for the
environment. In this context, the role of the regions as important units
for research, innovation and social cohesion has already been
acknowledged (European Commission, 2001). Therefore, finding means to
encourage economic growth at a regional level can help reduce regional
disparities while also boosting competitiveness for the whole Community.
For this purpose, we employ the concept of Total Factor
Productivity (TFP). TFP is the main driver of economic growth in most
mature economies, and understanding its determinants is essential in
devising policies that help enhance growth prospects and
competitiveness. The economic growth literature has shown theoretically
and empirically the importance of the Solow residual (TFP) over factor
accumulation. As a result, TFP which is mainly determined by
technological progress is the main determinant of the performance of a
country over time and also seems to account for the much of the
differences in income levels and the growth rates.
The recent literature also looks for additional determinants of
growth beyond the basic factors of production. A majority of these
growth empirics treat the determinants of output growth as inputs,
introducing them into the production function. However, these factors
may affect output growth indirectly, by affecting their efficiency
through an impact on production factors; this is the approach followed
here.
When it comes to TFP determinants, endogenous growth theory
establishes the role of innovative factors, such as R&D and human
capital, as important determinants of TFP formation, also recognising
that there are other factors that may lead to innovation and efficiency
improvements. In addition, the empirical literature suggests that
technological diffusion matters and thus, countries with low initial
levels of productivity can benefit from research and development
expenditures.
Even when it comes to these well-established determinants of TFP,
empirical studies seem to consider only one aspect, human capital is
usually proxied by education, and the health component is left aside.
However, health has been recognised as an important factor in growth
dating back to Grossman (1972) who modelled optimal investment in
increasing longevity. Generally speaking, improved health increases the
quality of the workforce, enhancing productivity and improving knowledge
absorption. Although a majority of studies focus on the health-growth
relationship in the underdeveloped countries, there are some arguments
that the health variable is also relevant in the context of developed or
developing economies. Tompa (2002) thinks that there are three channels
through which health impacts productivity: a healthy person has a higher
life expectancy, so she is keen to invest more in education, to save
more over the years encouraging capital accumulation and to have more
labour force participation. Also, the overall health status of the
population, as is the case with education, might attract or discourage
investments, especially FDI.
The role of infrastructure in stimulating output, efficiency and
productivity growth and also reducing production costs has been
considered in several empirical studies, since it draws the attention of
policy makers all over Europe. In the theoretical literature,
infrastructure is an important factor that can generate positive
external economies (see Romer (1986), Fucas (1988), and Barro and
Sala-i-Martin (1995)).
At the regional level, a labour force characterised by a high level
of human capital represents an advantage for the firms which could
enhance local productivity. The local economy benefits from healthy,
educated employees who attract foreign direct investors to implement
innovative activities inside the countries and thus, in fact, enhances
productivity for the whole economy. For regional TFP, we think that
R&D is important, since all regional firms may benefit from public
R&D and also from their own private R&D expenses. Recently, the
interest of theoreticians and practitioners from all over the world has
increased in what concerns the role of human capital in developing the
competitiveness of economies and, in particular, that of the regions.
Our paper brings additional evidence that human capital measured by
health has a positive influence on TFP growth at a regional level. We
also find evidence of R&D impact on regional productivity,
especially of public R&D in the developed EU15 regions. We find no
evidence of a significant private R&D effect; however, we argue that
private innovative activity is larger than what it is actually measured
by private R&D. Also, we share the opinion that the impact of public
R&D is reflected across the entire economy including the private
sector. This represents a good incentive for supporting further
investment in the health aspect of human capital and also in the public
innovation system. We also find evidence of an infrastructure effect on
TFP, although this relation is more stable in the regions of Central and
Eastern Europe (CEE) than in EU15.
The spatial dimension of regional analysis has also been discussed
in the context of competitive clusters (Porter, 2003) and spatial
spillovers (see Capello (2009) for a review on the matter). Innovation
and productivity tend to be 'clustered' in some specific areas
more than in others, but it could also be the case that productivity in
one region can be influenced by those of neighbouring regions. We show
that spatial conglomeration from capital cities matters for TFP creation
and R&D activity, especially in the EU15 regions. Moreover, most of
our results are robust when we control for the impact of neighbouring
regions' productivity, proving that our estimated effect of the TFP
determinants is not influenced by productivity spillover effects.
This paper is one of the few studies that attempts to study and
compare TFP determinants for both, EU15 and Central and Eastern Europe
(CEE) by considering R&D, health as a measure for human capital and
public infrastructure. Although there is a general consensus that
regions from newly acceded member states need to boost their
productivity and become more competitive, most of the studies (discussed
in the literature section) focus on the regions from developed EU
countries.
In our estimations, we use the levels of TFP computed from a
Cobb-Douglas production function with set input shares as the dependent
variable and we express all variables in logarithmic transformations. We
distinguish between public and private (business) R&D, as it is
known that the two can affect productivity in different ways and we
employ a novel variable that measures health: the number of doctors per
working age population. We use 123 NUTS regions, out of which 96 belong
to EU15 and 27 belong to CEE countries. Since EU15 and CEE regions have
different characteristics based on their economic conditions and
history, as well as differences in the length of their membership in the
Common Market, we assume they are characterised by different production
functions, so we present separate estimates for EU15 and CEE regions.
The time frame used is 1999-2010. In order to deal with endogeneity and
the dynamic character of the growth process, system GMM is used for
estimations.
The paper is organised as follows: the first section reviews the
literature on determinants of TFP, the next section discusses the
European Union regions analysed, the following section deals with the
data, variables and methodology, followed by a section that presents the
results and, finally, the conclusions.
LITERATURE REVIEW OF REGIONAL TFP DETERMINANTS
For the EU area, the growth accounting studies emphasise the
importance of TFP both in the western EU countries, considered to be
developed, and also in the recently added 11 CEE countries. Musso and
Westermann (2005) show that in Euro Area countries, the single most
important contributor to real GDP growth over 1980-2003 was TFP.
Schadler et al (2006) emphasise the importance of TFP to growth in CEE
countries during 1990-2004, stating that this large contribution of TFP
is what separates this group of countries from other emerging economies.
The studies that assess the impact of TFP determinants in the EU
regions focus mostly on the NUTS2 regions of the EU15 member states.
Ladu (2010) provides TFP estimates for 115 European regions over the
period 1976-2000, by using panel data cointegration techniques. Results
show that some regions of France and Austria have the highest
productivity, while the lowest TFP belongs to regions from Greece and
Spain. Bronzini and Piselli (2009) estimate the determinants of TFP for
the Italian regions over 1980-2001, by considering R&D, human
capital and public infrastructure. Their causality tests reveal that
there exists a long-run relationship between productivity level and the
three variables, the strongest relationship being between human capital
and TFP. Dettori et al. (2012) study the role played by intangible
factors on TFP creation, by analysing three types of capital: human
capital, social capital and technological capital, proxied by number of
patents. Their study also takes into account infrastructure, by
considering the region's accessibility by different means of
transport. Although they find evidence that all three types of capital
contribute to TFP formation, technological capital has the most
essential impact, being significant at 1% level in all specifications.
Vogel (2012) uses panel data from the manufacturing sector of 159 EU 15
regions and analyses both channels through which R&D and human
capital can affect TFP: directly, through innovation and indirectly,
through imitation. By allowing conditional convergence of TFP and
regional spillovers, their results prove that human capital has a
positive and direct effect on TFP, while R&D has a positive but
indirect effect.
The health variable, considered to contribute to productivity as
much as education, was first introduced in the growth models by Knowles
and Owen (1995), who augmented the Mankiw et al. (1992) model and found
a positive and significant relationship between health and economic
growth. More recently, Cooray (2013) investigates the impact of health
capital on economic growth disaggregated by income levels and finds that
in higher and upper middle countries, health has a positive and robust
influence on economic growth. Cole and Neumayer (2006) argue that a key
mechanism through which health affects growth is through TFP. In the
context of EU regions, there is no extensive study that considers the
explicit impact of health on productivity. When assessing the
productivity determinants in the Polish NUTS3 regions, Danska-Borsiak
and Laskowska (2012) construct a human capital index where, apart from
considering education and technology aspects (e.g. number of students,
percentage of graduates, Internet access), they also account for health
aspects, proxied by number of visits to physicians. Although their
results do show a positive impact of human capital on productivity, it
is difficult to assess from their estimates the specific effect of the
health variable on TFP.
PATTERNS OF THE EUROPEAN UNION REGIONS
This section describes the patterns in the evolution of TFP and its
possible determinants that emerge in both EU15 and CEE regions. As it
can be seen from Table 1, there are some important differences in the
average values between the regions from the two blocks of countries. As
expected, the differences in productivity are significant, with average
TFP levels in EU15 more than 5 times higher than those in the CEE. When
it comes to R&D intensity, EU15 regions also perform better,
dedicating 3.5 times more resources to research than the CEE regions;
however, their average intensity is still far from the EU target of 3 %.
The difference is especially significant when it comes to R&D
business, this sector is about one-fifth as large as a share of GDP in
the CEE as compared to the EU 15. It is important to note that in the EU
15 regions most of the R&D is carried out in the private sector,
while in the CEE regions the public sector has the leading role in
R&D activity. Even so, public sector R&D in the CEE regions is
less than one-half as large as in the EU15.
There are also major differences when it comes to the health
variable. The EU15 regions have around 1.3 doctors per working age
person, while in the CEE regions it is about one-third as large. The
differences in infrastructure, on the other hand, are very small, with
CEE and the EU15 regions having about the same amount of roads.
The plots from Figure 1 show a fairly linear relationship between
average productivity and the determinants we have considered. There
seems to be a positive relationship between TFP and total R&D
intensity. This trend is maintained also in the public and private
sector, with the relationship stronger (higher R-square) in the private
sector. The regions of the CEE distinguish themselves from the rest of
the regions, in the lower left corner of the scatter plots, as having
low productivity and low R&D intensity. The relationship between TFP
and health, although positive, seems to be rather weak when no other
factors are being considered.
The patterns presented here point towards a potential relationship
between TFP and the determinants we are looking at. Also, the major
differences between EU15 and CEE regions in terms of productivities,
R&D intensity and R&D structure confirm the need for considering
two different production functions. Further, we will estimate this
relationship in a panel data frame, considering various robustness
specifications.
[FIGURE 1 OMITTED]
DATA, VARIABLES AND METHODOLOGY
The econometric model
To estimate the determinants of TFP in the EU regions, the
following baseline equation was employed:
ln[A.sub.i,t], = [[beta].sub.0]ln[A.sub.i,t-1] + [[beta].sub.1] ln
[HC.sub.i,t] + [[beta].sub.2] ln [RD.sub.i,t] + [[beta].sub.3] ln
[INFR.sub.i,t] + [[eta].sub.i] + [[epsilon].sub.i,t] (1)
where [A.sub.i,t] is Total Factor Productivity, expressed in
levels, computed from Cobb-Douglas production function with constant
returns to scale, i.e. Y = [AK.sup.[alpha]] [L.sup.1-[alpha]], as
discussed below. Different capital share estimates are used for the EU15
([alpha] = 0.3) and CEE ([alpha] = 0.6) regions. We prefer a
parsimonious specification of the Cobb-Douglas production function as it
is a simple one and represents a good starting point in studying TFP,
dating back to Solow (1957). TFP expressed in ln can also be estimated
from the logarithmic form of the Cobb-Douglas production function;
however, this involves also estimating the input shares, which we have
done in previous research for the CEE countries and it is beyond the
scope of this paper. TFP can also be expressed in growth rates, easily
derived from a growth accounting exercise; however, using growth rates
instead of levels is considered to cause information loss in data and
hence, lead to less preferred estimates; [A.sub.i,t] is the lagged
dependent term, which shows the dynamic aspect of A. The fact that the
creation of new knowledge is based on the existing stock of knowledge
dates back to Romer (1990); [[beta].sub.0] may be interpreted as a
factor of conditional convergence and expresses the catch-up effect
towards the steady state. Conditional convergence, as explained under
the neoclassical growth theory, allows each region to have a different
level of productivity to converge towards; [HC.sub.i,t] is the human
capital variable, for which we used health as a proxy, measured by
number of doctors per working age population (15-64 years) (1);
[RD.sub.i,t] represents the total R&D investments, expressed as
percentages in GDP. Further on, we split R&D into business R&D
and public R&D, comprising government and educational R&D;
[INFR.sub.i,t] is a proxy for infrastructure, for which we used the
existent kilometres of road in the specific region; [[eta].sub.i] is the
region fixed effects that include time-invariant elements specific to a
region and [[epsilon].sub.i,t] represents the error term, assumed to be
homoscedastic and with no serial correlation.
We have also introduced time dummies variables, to reduce the
impact of time-specific effects across all regions and also to deal with
the persistence of the series. As a first robustness check, we control
for the regions which include the capital city, by introducing a country
capital dummy variable. Country capitals are usually the largest cities
in the country, and it is known that productivity tends to be higher in
large cities and the areas around them. As Harris and Moffat (2012)
point out, the diffusion and accumulation of knowledge is expected to be
better in areas with many people and this also creates a spatial
spillover effect that affects the surrounding region.
As mentioned, the level of TFP, the variable [A.sub.it] is computed
from a Cobb-Douglas specification, leading to the following equation:
A = [Y/L] x [(L/K).sup.[alpha]] (2)
where L is the labour stock measured by the working age population,
aged 15-64, K is the capital stock, computed with a Permanent Inventory
Method (PIM), using a depreciation rate of 5% for the CEE regions and 4%
for the EU15 regions. PIM is used to compute the capital stock from past
investment, depreciation and an estimate of an initial condition because
direct measurement of the stock of capital is practically impossible.
The depreciation rates chosen were based on the values used in the
literature (Nehru and Dhareshwar 1993), as well as on the general
assumption that the capital stock depreciates faster in developing
countries, as these countries are normally engaged in a growth process
based on capital accumulation and technology catching-up process.
-[alpha] is the capital input share, which was chosen based on the
existing literature for EU15 and our own estimates for the CEE. The
choice of capital share [alpha] = 0.6 for the CEE regions is based on
the authors' previous research results, see Pop Silaghi and Alexa
(2015) and is backed up by similar studies. In that study, we use
country level data for the 1993-2008 period and employing a
labour-augmented Cobb-Douglas production function in order to avoid an
over-inflated capital share. Our estimate of [alpha] = 0.6 is consistent
with other estimates for the CEE countries, see Iradian (2007) who also
finds capital shares between 0.4 and 0.78 for the Central and
South-Eastern Europe. A high value of alpha for the CEE regions makes
sense because the stock of capital includes foreign direct investment.
We did perform robustness checks by considering different values for
[alpha] in the 0.4-0.6 interval and obtained estimations that are in
line with the findings presented below.
To correct for possible endogeneity, we use a system GMM estimator.
This estimator was developed by Arellano and Bover (1995) and Blundell
and Bond (1998) building on the Arellano and Bond (1991) difference GMM.
By exploiting additional moment conditions, system GMM allows the use of
more instruments and hence, it is considered to be more efficient than
the difference GMM estimator. Intuitively, based on the assumption it
makes, the estimator permits the construction of a system of two
equations: the differenced one, where lags of the dependent and
independent variables are used as instruments, and the original one in
'levels' that use first differences as instruments. Besides
its improved econometric efficiency, system GMM is considered to be more
appropriate in growth empirics, as it could solve the problem of poor
instruments caused by high persistence of independent variable (Bond et
al., 2001). By instrumenting the independent variables, the problem of
endogeneity and double causality between the regressors and the
independent variable is also addressed. As we will see below, system GMM
also allows us to extend our model, by further considering spatial
dependence.
Implementing spatial dependence
Although our dummy variable for regions that include the capital
city captures some conglomeration and spatial effects, it does not
properly account for the spatial dependence that may arise in our model.
A majority of the growth models usually assume that the growth rates are
randomly distributed across spatial units; however, this hypothesis
might not be valid at a regional level. In recent years, following
developments in spatial econometrics, it has become standard to account
for spatial dependence in the context of regional growth. In our model,
in the case of innovation, there could be spillovers from highly
productive, highly innovative regions to surrounding regions which can
lead to biased estimates for different determinants on TFP. Not
accounting for spatial dependence works like an omitted variable bias
(LeSage and Pace, 2009, p. 27) as part of the estimated effects may be
in fact attributed to the geographical proximity between regions and not
to the actual correlations between variables. Intuitively, spatial
dependence can be introduced into the model in a nuisance form (spatial
error term models) or in a substantive form (spatial autoregressive
models), as stated by Anselin and Rey (1991). Although one would perform
tests to choose between the two models, it is customary for the growth
models to assume the second case, as it provides a meaningful
interpretation (Kubis and Schneider, 2012) and it proves to be the most
appropriate in a model of conditional convergence (Fingleton and
LopezBazo, 2006). Also, as Kubis and Schneider (2012) and Elhorst (2012)
advocate, neglecting the spatial dependence in a substantive form is
worse than neglecting the spatial autocorrelation in the error term, as
it affects the consistency of the estimator. Based on this, by using the
weight distance matrix W, we transform Equation (1) into a spatial
autoregressive lag model:
[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII], (3)
where Wln[A.sub.i,t] (represents the spatially lagged dependent
variable. The matrix W, which captures the effect of the interactions
between neighbouring regions, is computed in our case as the inverse
distance weighted matrix, illustrating the idea that the smaller the
distance between regions, the higher the spatial dependence between
them. To avoid unreasonable neighbourhood relationships over large
distances, we are considering that there is no spatial interaction
between regions that are more than 1000 km apart. The resulting matrix
is row-standardised, as is usual in the literature.
There are now a variety of estimators that deal with spatially
lagged variables in a data panel context; we use the approach of
Monteiro and Kukenova (2009) which is also used in other empirical
growth models (see Kubis and Schneider, (2012)). By means of Monte Carlo
simulations, Monteiro and Kukenova (2009) show that directly estimating
the System GMM with a spatially lagged dependent variable works
reasonably well, outperforming the alternative estimation strategies in
terms of bias and efficiency. We estimate Equation (3) in the same way
as Equation (1), treating WlnA as an endogenous term and instrumenting
it accordingly.
Data and model validity
For our estimations, we employed EUROSTAT data over the 1999-2010
period. In defining our geographical units of analysis, we follow
"Nomenclature of Statistical Territorial Units" NUTS
classification provided by EUROSTAT. We refer to the NUTS 2 regional
level, since these regions have their own administration. Due to data
availability, we use a limited number of the 123 NUTS regions, 96
belonging to EU15 and 27 to CEE countries. (2) The regions belong to 13
countries: Austria, Croatia, Finland, France, Germany, Italy,
Netherlands, Poland, Romania, Slovenia, Spain, Sweden and United
Kingdom. The choice of the regions included in our study (and therefore,
of the countries) is subject to data availability in the EUROSTAT
regional database. Different numbers of regions may appear in the
estimation tables, as the split between public and business R&D is
not available for all regions.
To avoid short-term business fluctuations affecting the results,
5-year rolling averages were used. Our panels are balanced, due to the
rolling averages used and each specification is tested for both EU15 and
CEE regions. As robustness checks, we employ a different time
frame--corresponding to the period before the 2004 EU enlargement--and
we consider the impact of spatial dependence in our model, as described
earlier.
In the case of system GMM, particular attention must be given to
the validity of the instruments used. Because of our sample size, we
could not employ all available lags as instruments, as one rule of thumb
is to keep the number of instruments smaller than the number of groups.
To achieve this, we have also used the 'collapsed' version of
our instruments, as recommended by Roodman (2009). The combination lags
used as instruments are described in the results (Tables 3, 4, 5),
although various lag combinations showing similar results were tested.
To test the validity of our instruments, we employed two tests: the
Hansen Test and the Difference-in-Hansen test. In our case, the p values
reported for Hansen Test are usually higher than 0.05, proving that both
sets of instruments--for the level and differenced equations--are fairly
valid. The Difference-in-Hansen test inquires the validity of the
"GMM-style" instruments for the levels equations, which should
be valid for the system GMM to be consistent. Again, the p values for
this test indicate that our system GMM is robust in most of the cases.
We also checked for stationarity in our data, as it is known that the
presence of unit roots in series can lead to spurious results. To test
for panel unit root processes, we employed a Fisher-type test which
performed an ADF unit root test on each panel and then used the inverse
normal transformation on the p values to obtain the overall test for the
panel series. This test is suitable for our T and N dimension and also
allows for different autoregressive parameters across the panels. The
inverse normal transformation we used was considered by Choi (2001) as
the most suitable, both in cases of finite and infinite N. The results
of the panel data unit root test, as described in Table 2, generally
reject the null hypothesis of random walk processes both in the case of
EU15 and CEE. Although in the case of infrastructure variable in EU15
the null hypothesis seems to be accepted, the p value is still small [p
= 0.118) so we do not consider that it represents a problem for our
model.
RESULTS
The estimates of Equation (1) for EU15 and CEE regions are
presented in Table 3. Column 1 shows the effect of health variable,
total R&D and infrastructure on TFP in the EU15 regions. Column 3
differentiates between business and public R&D in the process of TFP
formation. In columns (2) and (4), the impact of the region containing
the country capital is added to the specifications in columns (1) and
(3), respectively. The same four models are then re-estimated for the
CEE regions in columns (5)-(8). The results indicate that the number of
doctors per working age population seems to have a significant impact on
TFP in both EU15 and CEE regions; however, the effect is stronger and
more stable in the EU15 regions. Total R&D is significant both in
EU15 and CEE regions and when we separate between business and public,
only public R&D remains significant in the EU15 regions.
Infrastructure also appears to have an impact, especially in the CEE
regions.
The robustness check considering the period after the 2004 EU
enlargement indicates total R&D and public R&D as drivers of TFP
in the EU15 regions, whereas in the CEE regions infrastructure seems to
be a fairly robust driver of productivity (Table 4). The health variable
stays significant only in one equation for EU15 regions. It is known
that having access to a larger market and benefiting from unrestricted
trading of goods, people and ideas may have a stimulating effect on
productivity and innovation. However, this effect is not verified for
our CEE regions--it might take more time for the benefits of being part
of an economic union to pay off.
Including the country capital city affects the impact of R&D on
productivity in the EU15 regions, both in Table 3 and Table 4. The
effect of R&D decreases when we account for the region that includes
the capital of the country. This suggests that most of the innovative
R&D activity is absorbed in these country capital regions. This
impact is more robust with public R&D than it is with total R&D.
The result is expected since regions where the capital city is situated
are the most competitive regions in their respective countries (see
Annoni and Dijkstra (2013)); there is a conglomeration of human capital,
innovative companies and universities that boost productivity. Research
carried out in public institutions (universities, public institutions)
seems to be of particular interest here, as their effect on R&D is
shown to be significant and robust. However, for the sake of reducing
disparities, innovation measures should target other regions as well.
The results in Table 5 present an autoregressive spatial lag of the
dependent variable, taking into account the fact that the TFP in one
region may be influenced by that in a neighbouring region. We are aware
that in our model the spatial autocorrelation of the dependent variable
cannot be fully tested, as we are not employing all existing EU regions,
so the insignificance of [rho] (the coefficient on [WinA.sub.i,t])
should be interpreted carefully. The results do point out that spatial
proximity has an effect on productivity and also on the way in which
some factors contribute to the creation of TFP. These spatial effects
are less visible on the health variable, for which the coefficient stays
significant at the same level in the EU15 and CEE region samples. It can
be seen, however, that for the R&D variable, in most of the cases
its significance drops, as it is the case in columns (1), (2) and (6).
Especially when the effect of the capital city is considered, total
R&D becomes insignificant, both in EU15 and CEE. Public R&D
seems to be more robust to the spatial autocorrelation, while for the
EU15 its effect remains the same.
Out of all the considered determinants of TFP, the most robust seem
to be public R&D, whose impact is quite strong on the productivity
among the EU15 regions. Our results also point out that most of these
R&D activities are carried out in regions which include the country
capital, reinforcing the idea that innovation in EU regions is not a
homogenous process. R&D as a whole has a significant impact in the
regions of the EU15 and CEE; however, its effect is influenced by both
spatial dependence and conglomeration effect of the country capital.
From our results, nothing can be said about business R&D. It is true
that the innovation in the private sector is not always accounted by the
business R&D variable and this makes it difficult to properly
quantify the effect of firms' innovation on TFP. Therefore, our
results seem to support investments in public R&D in EU15 regions.
(3) For the CEE regions, although there is evidence of the total R&D
impact on TFP, the sector where it manifests itself is less clear. One
argument in favour of a significant impact of public R&D is also
enforced by the European Commission (2014), stating that the effects of
public R&D are seen across all the economy, even in the private
sector, as public R&D generates the knowledge base and talent that
private R&D needs.
Until now, health has been ignored as a determinant for total
productivities in the European regions. Although generally significant
and robust to spatial dependence, its effect fades after the second EU
enlargement. Infrastructure also has an effect that is especially
visible in the CEE regions. Infrastructure allows regions to be better
connected and to expand their productivity by engaging in commercial and
institutional exchanges. Especially in the case of CEE, this seems to be
a crucial factor, as historical and economic constraints have impeded
them to fully benefit from commercial transactions. From this point of
view, our paper encourages further investments in infrastructure.
Our results also point out the conditional convergence process that
it is taking place, in both EU15 and CEE, the regions are converging
towards their own steady-state productivity. Although this does not
represent an indicator that regional disparities are being reduced,
conditional convergence both within EU15 and CEE regions is a good sign,
as it may be 'a necessary (but not sufficient) condition of sigma
convergence' to take place (Young, et al., 2008). (4)
CONCLUSIONS
This paper assessed the impact of determinants of TFP across 96
EU15 regions and 27 CEE regions during 1999-2010, by employing variables
such as R&D, human capital and infrastructure. We not only looked at
the effect of total R&D intensity, but also disaggregated it by
sectors of activity, namely public R&D and private R&D. As a
measure for human capital, we used a health indicator, the number of
doctors per working age population in a given region. Based on the
recent literature, we also introduced an infrastructure variable,
proxied by the kilometres of roads in a given region. We also controlled
for spatial dependencies by introducing a spatial lag in our main
estimations. In order to eliminate any business cycle effects, we used
5-year rolling averages. We employed system GMM estimator, controlling
for possible endogeneity within our variables.
Our findings support the importance of R&D as a positive and
robust determinant of productivity, especially in the case of EU15. A
conditional catch-up process in terms of TFP seems to take place, and
this can continue if R&D is supported in the future, and also if
additional factors, ignored until now, such as health, are taken into
consideration. Our study brings additional evidence that human capital
proxied this time by a health dimension which seemed to us a suitable
proxy at regional level, has a positive influence on TFP.
Infrastructure, especially in CEE regions, counts for increasing total
factor productivity. While robustness checks are performed, taking into
account the year of the first enlargement of the European Union, health
and public R&D remain significant solely for the EU 15 regions. When
the capitals cities of each county are considered, the effect of R&D
decreases, however, with the public R&D effect being robust to the
impact of spatial autocorrelation.
Decreasing regional disparities is a challenge for policy makers,
so as to be able to emphasise policies that are suitable for all the EU
member countries. In order to do so, finding and knowing the factors
that can stimulate economic growth, in a direct way or indirectly by
positively influencing the total factor productivity, constitute a
priority. Considering three sources of productivity that are recognised
as potential drivers of growth, both at national and regional level, for
all regions of EU, based on data availability, represented both a
challenge and a desire to reveal some facts about regions inside EU and
their expected (or not) convergence. Finding a way through which health
is sustained at regional levels, no matter if the rural areas are
prevalent, should be a major concern of the policy makers. As we could
notice, health was significant for both EU 15 and CEE, proving that if
medical services are offered and thus, normally, the probability of
people being healthier if treated is higher. This positive aspect may be
seen in their performance at work, which means an increased level of
productivity. Finding channels of cooperation between public and private
sectors that invest in R&D could sustain long-term economic growth
rates at the regional levels and also high levels of total factor
productivities. Last, but not least, infrastructure, especially in CEE
regions, should be improved so as to permit these regions to develop
more, based on internal trading between them inside each country and
also based on better labour mobility that would contribute to reducing
disparities, firstly inside the country and secondly, between them and
other similar regions from neighbouring countries.
Acknowledgments
Monica Pop Silaghi carried the work for the final version of this
paper under the auspices of the grant for young researchers financed by
Babes-Bolyai University, GTC_34036.
REFERENCES
Annoni, P and Dijkstra, L. 2013. EU Regional Competitiveness Index
RCI 2013. JRC Scientific and Policy Reports: Luxembourg.
Anselin, L and Rey, S. 1991: Properties of tests for spatial
dependence in linear regression models. Geographical Analysis 23:
112-131.
Arellano, M and Bond, S. 1991: Some tests of specification for
panel data: Monte Carlo evidence and an application to unemployment
situation. The Review of Economic Studies 58: 277-297.
Arellano, M and Bover, 0. 1995: Another look at the instrumental
variable estimation of error-component models. Journal of Econometrics
68(1): 29-51.
Barro, R and Sala-i-Martin, X. 1995: Economic Growth. Second ed.
McGraw-Hill: New York.
Blundell, R and Bond, S. 1998: Initial conditions and moment
restrictions in dynamic panel data models. Journal of Econometrics 87:
115-143.
Bond, S, Hoeffler, A and Temple, J. 2001: GMM estimation of
empirical growth models. Working Papers, Nuffield College, University of
Oxford.
Bronzini, R and Piselli, P. 2009: Determinants of long-run regional
productivity: The role of R&D, human capital and public
infrastructure. Regional Science and Urban Economics 39(2): 187-199.
Capello, R. 2009: Spatial spillovers and regional growth: A
cognitive approach. European Planning Studies 17(5): 639-658.
Choi, I. 2001: Unit root tests for panel data. Journal of
International Money and Finance (20): 249-272.
Cole, MA and Neumayer, E. 2006: The impact of poor health on factor
productivity: An empirical investigation. Journal of Development Studies
42(6): 918-938.
Cooray, AV. 2013: Does health capital have differential effects on
economic growth? Applied Economics Letters 20(3): 244-249.
Dahska-Borsiak, B and Laskowska, 1. 2012: The determinants of total
factor productivity in Polish subregions: Panel data analysis.
Comparative Economic Research: Central and Eastern Europe 15: 17-29.
Dettori, B, Marrocu, E and Paci, R. 2012: Total factor
productivity, intangible assets and spatial dependence in the European
Region. Regional Studies 46(10): 1401-1416.
Elhorst, J. 2012: Dynamic spatial panels: Models, methods, and
inferences. Journal of Geographical Systems 14(1): 5-28.
European Commission. 2001: The Territorial Dimension of Research
and Development Policy: Regions in the European Research Area, s.l.:
Directorate--General for Research, European Commission.
European Commission. 2014. Research and Innovation as Sources of
Renewed Growth. Communication from the Commission to the European
Parliament, the Council, the European Economic and Social Committee and
the Committee of regions: Brussels.
Fingleton, B and Lopez-Bazo, E. 2006: Empirical growth models with
spatial effects. Papers in Regional Science 85(2): 177-198.
Grossman, M. 1972: On the concept of health capital and demand for
health. The Journal of Political Economy 80(2): 223-255.
Harris, R and Moffat, J. 2012: Is productivity higher in British
cities? Journal of Regional Science 52(5): 762-786.
Iradian, G. 2007: Rapid growth in transition economies:
Growth-accounting approach. IMF Working Paper WP/07/164, July.
Knowles, S and Owen, PD. 1995: Health capital and cross-country
variation in income per capita in the Mankiw-Romer-Weil Model. Economic
Letters 41(1): pp. 99-106.
Kubis, A and Schneider, L. 2012: Human capital mobility and
convergence. A spatial dynamic panel model of the German Regions.
IAB--Discussion Paper Volume 23.
Ladu, MG. 2010: Total factor productivity estimates: Some evidence
from European Regions. WIFO Working Papers, Issue 380.
LeSage, J and Pace, K. 2009: Introduction to Spatial Econometrics.
s.l.:Chapman and Hall/CRC: Boca Raton.
Lucas, R. 1988: On the mechanics of economic development. Journal
of Monetary Economics 22(1): 3-42.
Mankiw, G, Romer, D and Weil, D. 1992: A contribution to the
empirics of economic growth. The Quarterly Journal of Economics 107(2):
407-437.
Monteiro, J and Kukenova, M. 2009: Spatial dynamic panel model and
system GMM: A Monte Carlo investigation. Irene Working Papers 09-09,
IRENE Institute of Economic Research.
Musso, A and Westermann, T. 2005: Assessing Potential Output Growth
in the Euro Area: A growth accounting perspective. ECB Occasional Paper
No. 22.
Nehru, V and Dhareshwar, A. 1993: A new database on physical
capital stock: Sources, methodology and results. Revista de Analisis
Economico 8(1): 37-59.
Pop Silaghi, M and Alexa, D. 2015: Sources of growth: Evidence from
Ten Central and Eastern European Countries during 1993-2008.
Panoeconomicus 62(5): 643-661.
Pop Silaghi, M, Alexa, D, Jude, C and Litan, C. 2014: Do business
and public sector research and development expenditures contribute to
economic growth in Central and Eastern European countries? A dynamic
panel estimation. Economic Modelling 36: 108-119.
Porter, M. 2003: The economic performance of the regions. Regional
Studies 37(6-7): 549-578.
Romer, P. 1986: Increasing returns and long term growth. The
Journal of Political Economy 94(5): 1002-1037.
Romer, P. 1990: Endogenous technological change. Journal of
Political Economy 98(5): s71-sl02.
Roodman, D. 2009: How to do xtabond2: An introduction to difference
and system GMM in Stata. The Stata Journal 9(1): 86-136.
Schadler, S, Mody, A, Abiad, A and Leigh, D. 2006: Growth in the
Central and Eastern European Countries of the European Union.
International Monetary Fund, Occasional Paper no. 252.
Solow, RM. 1957: Technical change and the aggregate production
function. The Review of Economics and Statistics 39(3): 312-320.
Tompa, E. 2002: The impact of health on productivity: Empirical
evidence and policy implications. The Review of Economic Perfomance and
Social Progress, Volume 2.
Vogel, J. 2012: The two faces of R&D and Human Capital:
Evidence from Western European Region. Discussion Paper Series, no. 599.
Young, A, Higgins, M and Levy, D. 2008: Sigma convergence versus
beta convergence: Evidence from US County Level Data. Journal of Money,
Credit and Banking, 40(5): 1083-1093.
(1) Other proxies for Human capital are also employed, such as life
expectancy or health expressed as absolute number of doctors. However,
no convincing results were obtained. We preferred to use the component
of human capital that relates to health rather than education, at
regional level.
(2) The complete list of the NUTS2 regions used in estimations can
be obtained from the authors.
(3) In an earlier paper (Pop-Silaghi et ai, 2014), we assessed the
impact of private and public R&D on the economic growth rates at a
national level, just for CEE countries. Our results showed that private
R&D was significant and its effect was robust in many
specifications.
(4) For the CEE regions, the results obtained in Table 3 for column
(5) and (8) remain fairly robust for different capital share
specifications (alpha between 0.45 and 0.60), confirming one of the main
findings of this paper for the CEE: over the studied period, in the CEE
regions, total R&D, health and infrastructure have positive impact
on TFP. These results are available upon request.
DIANA ALEXA [1,2], MONICA POP-SILAGHI [3] & LAURA MARIANA
CISMAS [1]
[1] Faculty of Economics and Business Administration, West
University of Timisoara, Timisoara, Romania. E-mail:
dianaalexams@gmail.com
[2] 58, Ridgepool Village, Ballina, Co. Mayo, Ireland.
[3] Faculty of Economics and Business Administration,
Babe[section]-Bolyai University, Cluj-Napoca, Romania.
This symposium paper was presented at the XIIIth edition of the
International Finance and Banking (FIBA) Conference organized by the
Faculty of Finance of the Bucharest School of Economic Studies which was
held on March 26-27, 2015 in Bucharest, Romania.
Table 1: Average values of possible TFP determinants
EU15 CEE Total
TFP levels 11.71 2.15 9.76
R&D total intensity (% in GDP) 1.54 0.44 1.32
R&D business intensity (% in GDP) 0.91 0.17 0.76
R&D public intensity (% in GDP) 0.63 0.29 0.57
Health (No. of doctors per working 0.47 1.31 1.07
age population 15-64)
Infrastructure (km of roads) 18022 18879 18210
Note: R-square values are 0.23 for (TFP-R&D total); 0.14 for
(TFP-R&D public); 0.22 for (TFP-R&D business); 0.03 for (A-no
of doctors 15-64).
Source: Own computations based on EUROSTAT (2013) data
Table 2: Panel unit root test results
EU15 CEE
Z-statistic p Value Z-statistic p Value
ln A -11.630 0.000 -1.683 0.046
lnhealthdoctors15_64 -14.201 0.000 -4.031 0.000
lnrdtotal -8.045 0.000 -4.231 0.000
lnrdbusiness -8.020 0.000 -5.833 0.000
lnrdpublic -8.019 0.000 -5.971 0.000
lnroads -1.186 0.118 -6.574 0.000
A Fisher-type panel unit root test was employed, based on the AOF
tests for each panel. Panel-specific AR term was considered, with
fixed effects, drift term and 1st lag for the ADF regressions.
Inverse normal Z-statistic was reported, suitable for both finite
and infinite N. The test assumes null hypothesis Ho: All panels
contain unit roots, with the alternative Ha: At least one panel
is stationary.
Source: Own estimations based on EUROSTAT (2013) data
Table 3: Determinants of total factor
productivity in the European regions
(1) (2) (3)
EU15 EU15 EU15
[lnA.sub.i,t-1] 0.837 *** 0.811 *** 0.785 ***
(0.074) (0.091) (0.097)
[nHC.sub.i,t 0.068 ** 0.063 * 0.073 **
(0.030) (0.037) (0.029)
[lnRDtotal.sub.i,t] 0.057 *** 0.049 **
(0.016) (0.021)
[lnRDbusiness.sub.i,t] 0.009
(0.031)
[lnRDpublic.sub.i,t] 0.126 **
(0.049)
[lnINFR.sub.i,t] 0.040 0.037 0.028
(0.027) (0.030) (0.021)
[Countrycapital.sub.i,t] 0.109
(0.269)
No of regions 96.000 96.000 89.000
No of obs 659.000 659.000 579.000
Hansen Test (a) 0.114 0.134 0.464
Diff in Hansen (a) 0.502 0.436 0.274
AR(2) (a) 0.086 0.059 0.143
No of instruments 25.000 25.000 27.000
(4) (5) (6)
EU15 CEE CEE
[lnA.sub.i,t-1] 0.760 *** 0.939 *** 0.900 ***
(0.115) (0.034) (0.048)
[nHC.sub.i,t 0.056 0.076 ** 0.064 *
(0.065) (0.029) (0.035)
[lnRDtotal.sub.i,t] 0.051 ** 0.048 **
(0.021) (0.019)
[lnRDbusiness.sub.i,t] 0.000
(0.025)
[lnRDpublic.sub.i,t] 0.106 *
(0.063)
[lnINFR.sub.i,t] 0.032 0.065 *** 0.082 ***
(0.029) (0.017) (0.027)
[Countrycapital.sub.i,t] 0.295 0.153
(0.372) (0.121)
No of regions 89.000 27.000 27.000
No of obs 579.000 171.000 171.000
Hansen Test (a) 0.687 0.203 0.429
Diff in Hansen (a) 0.628 0.242 0.289
AR(2) (a) 0.312 0.288 0.099
No of instruments 27.000 25.000 25.000
(7) (8)
CEE CEE
[lnA.sub.i,t-1] 0.983 *** 0.982 ***
(0.090) (0.063)
[nHC.sub.i,t 0.038 0.031
(0.052) (0.058)
[lnRDtotal.sub.i,t]
[lnRDbusiness.sub.i,t] 0.022 0.031
(0.056) (0.054)
[lnRDpublic.sub.i,t] 0.051 0.059
(0.048) (0.039)
[lnINFR.sub.i,t] 0.034 0.033
(0.053) (0.044)
[Countrycapital.sub.i,t] -0.080
(0.077)
No of regions 23.000 23.000
No of obs 128.000 128.000
Hansen Test (a) 0.474 0.461
Diff in Hansen (a) 0.622 0.605
AR(2) (a) 0.245 0.239
No of instruments 27.000 27.000
Note: Standard errors in brackets. *, ** and *** denote
significance levels of 10, 5 and 1 %, respectively. Method used
is system GMM, with robust standard error, consistent with
panel-specific autocorrelation and heteroskedasticity in two-step
estimation. All regressions include time fixed effects.
Instruments for the first-difference equation: the 4th up to
the 7th lag of [lnA.sub.i,t-1], the 3rd and 4th lag of
[lnHC.sub.i,t], [lnRDtotal.sub.i,t], [lnRDbusiness.sub.i,t],
[lnINFR.sub.i,t] (collapsed), the 4th lag of [lnRDpublic.sub.i,t]
(collapsed).
Instruments for the levels equation: the first difference of the
lagged dependent and independent variables. Different lags
combinations were also tested for the first-difference equation.
(a) p values are reported.
The capital share used for computing TFP was 0.3 for EU15 regions
and 0.6 for the CEE regions.
Source: Own estimations based on EUROSTAT (2013) data
Table 4: Robustness check: determinants of total factor
productivity in the European regions after the 2004 EU
enlargement
(1) (2) (3)
EU15 EU15 EU15
[lnA.sub.i,t-1] 0.854 *** 0.845 *** 0.741 ***
(0.079) (0.083) (0.045)
[lnHC.sub.i,t] 0.037 0.049 0.066 **
(0.028) (0.047) (0.032)
[lnRDtotal.sub.i,t] 0.059 *** 0.053 ***
(0.017) (0.019)
[lnRDbusiness.sub.i,t] 0.010
(0.018)
[lnRDpublic.sub.i,t] 0.114 **
(0.046)
[lnINFR.sub.i,t] 0.010 0.018 0.042 **
(0.032) (0.028) (0.020)
Countrycapital.sub.i,t] 0.091
(0.300)
No of groups 95 96 89
No of obs 568 558 495
Hansen Test (a) 0.238 0.203 0.579
Diffin Hansen (a) 0.134 0.153 0.372
AR(2) (a) 0.281 0.240 0.393
No of instruments 20 20 23
(4) (5) (6)
EU15 CEE CEE
[lnA.sub.i,t-1] 0.728 *** 0.992 *** 0.957 ***
(0.051) (0.053) (0.070)
[lnHC.sub.i,t] 0.073 0.036 0.049
(0.067) (0.049) (0.054)
[lnRDtotal.sub.i,t] 0.028 0.029
(0.042) (0.039)
[lnRDbusiness.sub.i,t] 0.001
(0.027)
[lnRDpublic.sub.i,t] 0.110 *
(0.060)
[lnINFR.sub.i,t] 0.044 ** 0.070 ** 0.073 **
(0.020) (0.033) (0.033)
Countrycapital.sub.i,t] 0.242 0.079
(0.449) (0.132)
No of groups 89 27 27
No of obs 495 148 148
Hansen Test (a) 0.560 0.108 0.068
Diffin Hansen (a) 0.412 0.118 0.020
AR(2) (a) 0.520 0.128 0.279
No of instruments 23 20 20
(7) (8)
CEE CEE
[lnA.sub.i,t-1] 0.983 *** 0.993 ***
(0.084) (0.080)
[lnHC.sub.i,t] 0.034 0.002
(0.102) (0.092)
[lnRDtotal.sub.i,t]
[lnRDbusiness.sub.i,t] 0.021 0.039
(0.080) (0.059)
[lnRDpublic.sub.i,t] 0.034 0.054
(0.033) (0.037)
[lnINFR.sub.i,t] 0.041 0.020
(0.041) (0.047)
Countrycapital.sub.i,t] -0.070
(0.082)
No of groups 23 23
No of obs 113 113
Hansen Test (a) 0.168 0.095
Diffin Hansen (a) 0.122 0.051
AR(2) (a) 0.362 0.382
No of instruments 23 23
Note: Standard errors in brackets. *, ** and *** denote
significance levels of 10, 5 and 1%, respectively.
Method used is system GMM, with robust standard error, consistent
with panel-specific autocorrelation and heteroskedasticity in
two-step estimation. All regressions include time fixed effects.
Instruments for the first-difference equation: the 4th up to the
7th lag of [lnA.sub.i,t-1], the 3rd and 4th lag of
[lnHC.sub.i,t], [lnRDtotal.sub.i,t], [lnRDbusiness.sub.i,t],
[lnINFR.sub.i,t] (collapsed), the 3rd and 4th lag of
lnRDpublici/t (collapsed).
Instruments for the levels equation: the first difference of the
lagged dependent and independent variables. Different lags
combinations were also tested for the first-difference equation.
(a) p values are reported.
The capital share used for computing TFP was 0.3 for EU15 regions
and 0.6 for the CEE regions.
Source: Own estimations based on EUROSTAT (2013) data
Table 5: Robustness check: accounting for spatial dependence
when assessing determinants of TFP in European regions
(1) (2) (3)
lnA lnA lnA
b/se b/se b/se
[lnA.sub.i,t-1] 0.798 *** 0.764 *** 0.821 ***
(0.141) (0.208) (0.073)
[WlnA.sub.i,t] -0.066 -0.055 0.032
(0.131) (0.075) (0.074)
[lnHC.sub.i,t] 0.062 ** 0.064 * 0.080 ***
(0.029) (0.034) (0.022)
[lnRDtotal.sub.i,t] 0.058 ** 0.056
(0.023) (0.038)
[lnRDbusiness.sub.i,t] 0.011
(0.018)
[lnRDpublic.sub.i,t] 0.098 **
(0.041)
[lnINFR.sub.i,t] 0.040 * 0.048 0.039 **
(0.023) (0.040) (0.016)
[Countrycapital.sub.i,t] 0.074
(0.292)
No of groups 96 96 89
No of obs 659 659 579
Hansen Test 0.190 0.216 0.218
Difference-in-Hansen 0.535 0.353 0.225
AR(2) 0.039 0.041 0.029
No of instruments 27 27 29
(4) (5) (6)
lnA lnA lnA
b/se b/se b/se
[lnA.sub.i,t-1] 0.795 *** 0.928 *** 0.900 ***
(0.087) (0.035) (0.051)
[WlnA.sub.i,t] 0.007 0.037 0.133
(0.054) (0.146) (0.157)
[lnHC.sub.i,t] 0.064 * 0.069 ** 0.055 *
(0.032) (0.032) (0.029)
[lnRDtotal.sub.i,t] 0.060 *** 0.031
(0.018) (0.019)
[lnRDbusiness.sub.i,t] 0.014
(0.017)
[lnRDpublic.sub.i,t] 0.096 *
(0.048)
[lnINFR.sub.i,t] 0.033 0.058 *** 0.078 *
(0.023) (0.017) (0.040)
[Countrycapital.sub.i,t] -0.067 0.188
(0.212) (0.168)
No of groups 89 27 27
No of obs 579 171 171
Hansen Test 0.291 0.234 0.286
Difference-in-Hansen 0.308 0.290 0.120
AR(2) 0.050 0.239 0.127
No of instruments 29 27 27
(7) (8)
lnA lnA
b/se b/se
[lnA.sub.i,t-1] 0.955 0.961 *
(5.232) (0.558)
[WlnA.sub.i,t] 0.220 0.229
(18.630) (1.832)
[lnHC.sub.i,t] -0.009 0.016
(4.411) (0.289)
[lnRDtotal.sub.i,t]
[lnRDbusiness.sub.i,t] 0.023 0.013
(3.866) (0.273)
[lnRDpublic.sub.i,t] 0.050 0.042
(1.740) (0.304)
[lnINFR.sub.i,t] 0.015 0.026
(2.324) (0.301)
[Countrycapital.sub.i,t] 0.057
(1.316)
No of groups 23 23
No of obs 128 128
Hansen Test 0.682 0.696
Difference-in-Hansen 0.413 0.822
AR(2) 0.973 0.824
No of instruments 29 29
Note: Standard errors in brackets. *, ** and *** denote
significance levels of 10, 5 and 1%, respectively.
Method used is system GMM, with robust standard error, consistent
with panel-specific autocorrelation and heteroskedasticity in
two-step estimation.
All regressions include time dummies fixed effects.
Instruments for the first-difference equation: the 4th up to the
7th lag of [lnA.sub.i,t-1], the 3rd and 4th lag of [lnHC.sub.i,t],
[lnRDtotal.sub.i,t], [lnRDbusiness.sub.i,t], [lnINFR.sub.i,t]
(collapsed), the 4th lag of [lnRDpublic.sub.i,t] (collapsed), the
4th lag of [WlnA.sub.i,t] (collapsed).
Instruments for the levels equation: the first difference of the
lagged dependent and independent variables. Different lags
combinations were also tested for the first-difference equation.
(a) p values are reported.
The capital share used for computing TFP was 0.3 for EU15 regions
and 0.6 for the CEE regions.
Source: Own estimations based on EUROSTAT (2013) data.
